GNSS-denied geolocalization of UAVs using terrain-weighted constraint optimization

Accurate geolocation using Global Navigation Satellite Systems (GNSS) is essential for safe and long-range unmanned aerial vehicles (UAVs) flights. However, GNSS systems are susceptible to blockages, jamming, and spoofing attacks. Localization using onboard cameras and satellite images provides a promising solution for UAVs operating in GNSS-denied environments. In this paper, we developed a novel UAV visual localization system for GNSS-denied situations, both day and night, that integrates image matching, visual odometry (VO), and terrain-weighted constraint optimization. First, an effective map management strategy is designed for satellite image chunking, real-time scheduling, and merging. Then, a 2D–3D geo-registration method, combining Bidirectional Homologous Points Search, is introduced to obtain accurate 3D virtual control points for UAV absolute localization. Lastly, a position estimation and optimization method, integrating the sliding window with terrain weighting constraints, is proposed to control position error accumulation and reduce position drift. Twenty experiments were conducted in typical and complex scenarios to validate our system’s resilience to altitude changes, trajectory variations, and rolling terrain. Our system demonstrated drift-free and viewpoint-robust, maintaining stability even in feature-poor environments and seasonal variations. It does not require loop closure, allowing for re-localization after positioning failures. Additionally, we utilized thermal infrared images to demonstrate the system’s performance in night-time conditions. With a Mean Absolute Error of less than 7 m, it can be a powerful complement to GNSS in the event of GNSS-Denied environments. All demonstration videos of our system can be found at https://github.com/YFS90/GNSS-Denied-UAV-Geolocalization . • A UAV visual localization system for GNSS-denied situations both day and night. • BHPS and 2D–3D Geo-Registration to extract stable virtual control points. • Terrain-weighted constraint optimization greatly improves UAV positioning. • Twenty data sets validate the system’s adaptability and performance.